Independent restored v7 ALE review

LOCAL review of music-quintet-v7-20261009. Reviewed 2026-10-09T18:57:42.858270+00:00.

Scope and frozen identity

/apdcephfs_fsgm/share_304220499/nlphuang/ale_hardening_1494_storage/music_exp2021/frozen_v7_submission_restored

45 files; 2048309883 bytes. Inventory fingerprint cef4e48725c4aea346b9b7b99b94f307442714f658058ae4d4106dc6f4b1a0a3. Only this payload supplied task facts. The supplied September 29 public-criteria snapshot was used; official sources were not refreshed.

Same-filesystem read alias was independently checked by inode, size, mtime and sampled SHA, then full reference hashes were checked by the untouched CLI. Only controls memoize repeated trusted-file SHA reads while identity and timestamps match; this does not alter the full reference grader or musical scoring functions.

Reference evidence

{
  "version": "music-quintet-v7-20261009",
  "official_result": false,
  "components": {
    "pitch": 1.0,
    "rhythm": 1.0,
    "dynamics": 1.0,
    "instrument": 1.0,
    "timbre": 1.0,
    "arrangement": 1.0
  },
  "errors": [],
  "reference_notes": 20965,
  "candidate_notes": 20965,
  "native_replay_directory": "/mnt/wzw_probe/apdcephfs_fsgm/share_304220499/nlphuang/ale_hardening_1494_storage/music_v7_restored_payload_independent_review/full_reference_work/native-bc0721d55cbd",
  "native_contract_valid": true,
  "native_contract_violations": {},
  "native_export_channel_map": [
    {
      "role": "piano",
      "native_channel": 0,
      "export_channel": 0
    },
    {
      "role": "violin",
      "native_channel": 1,
      "export_channel": 1
    },
    {
      "role": "viola",
      "native_channel": 2,
      "export_channel": 2
    },
    {
      "role": "cello",
      "native_channel": 3,
      "export_channel": 3
    },
    {
      "role": "double_bass",
      "native_channel": 4,
      "export_channel": 4
    }
  ],
  "native_controller_export_agreement": 1.0,
  "native_note_export_agreement": 1.0,
  "parts": [
    {
      "role": "piano",
      "note_f1": 1.0,
      "native_render_agreement": 1.0,
      "dynamic_contour": 1.0,
      "timbre_shape": 1.0
    },
    {
      "role": "violin",
      "note_f1": 1.0,
      "native_render_agreement": 1.0,
      "dynamic_contour": 1.0,
      "timbre_shape": 1.0
    },
    {
      "role": "viola",
      "note_f1": 1.0,
      "native_render_agreement": 1.0,
      "dynamic_contour": 1.0,
      "timbre_shape": 1.0
    },
    {
      "role": "cello",
      "note_f1": 1.0,
      "native_render_agreement": 1.0,
      "dynamic_contour": 1.0,
      "timbre_shape": 1.0
    },
    {
      "role": "double_bass",
      "note_f1": 1.0,
      "native_render_agreement": 1.0,
      "dynamic_contour": 1.0,
      "timbre_shape": 1.0
    }
  ],
  "mix_native_agreement": 1.0,
  "stem_sum_agreement": 1.0,
  "score": 0.9999999999999999,
  "evidence_only": {
    "overview.png": true,
    "qc_note.txt": true
  }
}

Criteria

ID	Status	Finding	Evidence
A1	PASS	MusicNet primary release, source metadata and published annotation procedure provide a clear identified research/workflow source for a disclosed reconstruction task. This is a simulated brief using real recordings; independent musical QC and professional calibration are separately pending.	`workspace/input/ATTRIBUTION.txt:1; provenance/musicnet_record.json:13`
A2	PASS	One continuous 1072.176-second reconstruction across three complete movements and five roles; editable native project, MIDI, master, stems, regions and QC evidence form an end-to-end workflow. Expert elapsed time is unmeasured.	`workspace/instruction.md:1`
A3	PASS	Real concatenated audio, exact movement boundaries, simulated delivery brief and constructed dynamics are disclosed; packaged source labels and shifted combined labels agree.	`workspace/input/movement_manifest.json:1; controls.json`
A4	PASS	Recorded REAPER, LV2 and sample hashes match actual replay dependencies; Docker replay succeeds using the recorded software paths. This is the acquired environment, not a newly provisioned clean install.	`workspace/software.md:1; full_reference_grade.json`
A5	PENDING	Weights, matching rules, accepted equivalents, partial credit and error semantics are public and implemented. Musician calibration of author-selected dynamic/tone thresholds remains pending.	`workspace/input/delivery_contract.md:1`
A6	PASS	Actual fresh native reopening, complete note inspection/rendering and unmodified six-component grading independently reproduce the delivered reference. CSV-to-truth checking establishes annotation consistency, not perfect acoustic truth.	`full_reference_grade.json; controls.json; provenance/current_native_execution.json:1`
A7	PENDING	CC-BY-4.0 primary metadata and source attribution are included; GeneralUser licence is included and REAPER evaluation/licence conditions are disclosed. Final contributor and recording-specific distribution confirmation is explicitly pending.	`workspace/input/ATTRIBUTION.txt:1; provenance/generaluser_LICENSE.txt:1; workspace/software.md:1`
A8	NA	Other tasks and versions were excluded by review scope; the three movements are one workflow, not three independent contributions.	`scope.json`
A9	PENDING	Annotation-assisted deterministic reference, mixed-envelope velocities, real recording and simulated brief are disclosed. No commissioned-human QC or actual special-screening service result is supplied. Technical self-consistency does not establish source eligibility.	`README.txt:1; reference/qc_note.txt:1; builders/rebuild.py:1`
A10	PASS	All frozen payload files are inventoried and SHA-256 readable; screenshot was visually inspected; no nested or encrypted archives are present. Native runtime is separately exercised.	`before_inventory.json; full_reference_grade.json`
T1	PASS	Only workspace is solver-visible. Hidden source annotations, truth, builders, reference and evaluator are outside it. Export helper contains no transcription or labels.	`README.txt:1; before_inventory.json; workspace/software/export_native.py:1`
T2	PASS	Missing/bad/symlink MIDI gets candidate zero; absent truth, changed inspection worker and unavailable grader work path raise their respective null-score errors. Missing native/audio components retain the public 0.6 MIDI-only credit.	`controls.json; missing_reference.json; environment_error.json; unavailable_work.json; midi_only.json`
T3	PASS	Music is matched from MIDI events; native notes/controllers are read by actual REAPER; exports are compared to replayed audio. Screenshot and self-report fields do not contribute scores.	`evaluation/evaluate.py:1; evaluation/native_inspect.lua:1`
T4	PASS	Bounded controls cover invalid deliveries, wrong pitch/timing, duplicate notes, silent/nonfinite/full-scale audio, channel-controller association, native master gain, disabled direct master routing, cropped hidden notes and unsafe native media. This is not exhaustive adversarial-code certification.	`controls.json; native_controls.json`
T5	PASS	Complete-note PPQ, tempo and consistent channel renames retain agreement; duplicate absolute controller declarations collapse while reset commands and meaningful state restoration remain. Fresh full replay agrees with stored audio. Alternative instrument families were not professionally calibrated.	`controls.json; full_reference_grade.json`
T6	PENDING	Candidate delivery is replayed from a fresh relocated copy2 directory. Package rebuild uses packaged labels, but a new clean installation/provisioning and full author rebuild were not executed; generic acquired software remains at recorded paths and environment overrides are available.	`full_reference_command.json; evaluation/native_replay.py:1; builders/rebuild.py:1`
T7	PASS	Package explicitly describes a standalone contributor rubric and makes no executed remote ALE integration claim.	`README.txt:1`
D1	PENDING	No strong-agent trial or model trajectory was read or requested in this independent review. No official near-term or last-exam difficulty qualification follows from reference/control scores.	`scope.json`
S1	NOT RUN	No authenticated official task-fit request or result was supplied or executed.	`scope.json`
S2	NOT RUN	No official submission, receipt or acceptance was supplied or performed.	`scope.json`

Executed controls

Control	Status
truth_matches_packaged_source_annotations	PASS
all_source_labels_retained_and_offset	PASS
input_duration_and_silent_gaps	PASS
missing_output	PASS
missing_manifest	PASS
malformed_midi	PASS
symlink_midi	PASS
missing_reference_is_error	PASS
changed_worker_is_environment_error	PASS
unavailable_work_is_environment_error	PASS
missing_optional_components_retain_midi_credit	PASS
ppq_and_channel_rename_equivalent	PASS
native_export_channel_rename_equivalent	PASS
tempo_and_ppq_equivalent	PASS
wrong_pitch_loses_pitch_credit	PASS
duplicates_penalized	PASS
timing_shift_penalized	PASS
zero_audio	PASS
nan_audio	PASS
inf_audio	PASS
over_scale_audio	PASS
absolute_controller_duplicate_equivalent	PASS
controller_commands_not_collapsed	PASS
reset_sensitive_state_restoration_retained	PASS
reset_preserved_duplicate_collapsed	PASS
parameter_data_entry_context_retained	PASS
controller_wrong_voice_penalized	PASS
master_gain_native_rejected	PASS
direct_master_disabled_native_rejected	PASS
hidden_item_notes_native_rejected	PASS
absolute_media_rejected	PASS
missing_media_rejected	PASS

Pending evidence

Professional musical QC and dynamic/timbral calibration
Final contributor and source-recording distribution/rights confirmation
Strong-model difficulty qualification and expert-time evidence if claimed
Clean-environment provisioning/full rebuild validation if claimed
Official screening, task-fit, submission and acceptance

Limits

Research annotations carry the disclosed upstream approximately 4% label uncertainty.
Constructed dynamics are not independently observed per-instrument acoustic truth.
Passing six-component self-grade establishes executed consistency, not expert-approved transcription or official acceptance.
Tests are bounded to this payload; no previous versions, historical reviews, model trajectories, author cases or operational directories were used as task evidence.

Evidence files: review.json, full_reference_grade.json, controls.json, before_inventory.json, after_inventory.json, alias_identity.json, control_integrity_cache.json, programs/, full_reference.log and controls_complete.log. Bulk native execution receipts and WAVs remain under the review fixture root. The older controls.log is a retired partial input-check run with no score.